index_destroy() must grab exclusive access to the parent table

of the index it wants to destroy.  This ensures that no other backend is
actively scanning or updating that index.  Getting exclusive access on
the index alone is NOT sufficient, because the executor is rather
cavalier about getting locks on indexes --- see ExecOpenIndices().
It might be better to grab index locks in the executor, but I'm not
sure the extra lockmanager traffic is really worth it just to make
index_destroy cleaner.

src/backend/catalog/index.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/catalog/index.c,v 1.95 1999/11/16 04:13:55 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/catalog/index.c,v 1.96 1999/11/21 20:01:10 tgl Exp $
  *
  *
  * INTERFACE ROUTINES
@@ -72,6 +72,7 @@ static void DefaultBuild(Relation heapRelation, Relation indexRelation,
 		int numberOfAttributes, AttrNumber *attributeNumber,
 		IndexStrategy indexStrategy, uint16 parameterCount,
 		Datum *parameter, FuncIndexInfoPtr funcInfo, PredInfo *predInfo);
+static Oid IndexGetRelation(Oid indexId);
 
 /* ----------------------------------------------------------------
  *	  sysatts is a structure containing attribute tuple forms
@@ -1109,7 +1110,8 @@ index_create(char *heapRelationName,
 void
 index_destroy(Oid indexId)
 {
-	Relation	userindexRelation;
+	Relation	userHeapRelation;
+	Relation	userIndexRelation;
 	Relation	indexRelation;
 	Relation	relationRelation;
 	Relation	attributeRelation;
@@ -1118,12 +1120,21 @@ index_destroy(Oid indexId)
 
 	Assert(OidIsValid(indexId));
 
-	userindexRelation = index_open(indexId);
-
-	/*
-	 * Get exclusive lock to ensure no one else is scanning this index.
+	/* ----------------
+	 *	To drop an index safely, we must grab exclusive lock on its parent
+	 *	table; otherwise there could be other backends using the index!
+	 *	Exclusive lock on the index alone is insufficient because the index
+	 *	access routines are a little slipshod about obtaining adequate locking
+	 *	(see ExecOpenIndices()).  We do grab exclusive lock on the index too,
+	 *	just to be safe.  Both locks must be held till end of transaction,
+	 *	else other backends will still see this index in pg_index.
+	 * ----------------
 	 */
-	LockRelation(userindexRelation, AccessExclusiveLock);
+	userHeapRelation = heap_open(IndexGetRelation(indexId),
+								 AccessExclusiveLock);
+
+	userIndexRelation = index_open(indexId);
+	LockRelation(userIndexRelation, AccessExclusiveLock);
 
 	/* ----------------
 	 *	DROP INDEX within a transaction block is dangerous, because
@@ -1137,14 +1148,13 @@ index_destroy(Oid indexId)
 	 *	they don't exist anyway.  So, no warning in that case.
 	 * ----------------
 	 */
-	if (IsTransactionBlock() && ! userindexRelation->rd_myxactonly)
+	if (IsTransactionBlock() && ! userIndexRelation->rd_myxactonly)
 		elog(NOTICE, "Caution: DROP INDEX cannot be rolled back, so don't abort now");
 
 	/* ----------------
 	 * fix DESCRIPTION relation
 	 * ----------------
 	 */
-
 	DeleteComments(indexId);
 	
 	/* ----------------
@@ -1198,14 +1208,18 @@ index_destroy(Oid indexId)
 	heap_close(indexRelation, RowExclusiveLock);
 
 	/*
-	 * flush cache and physically remove the file
+	 * flush buffer cache and physically remove the file
 	 */
-	ReleaseRelationBuffers(userindexRelation);
+	ReleaseRelationBuffers(userIndexRelation);
 
-	if (smgrunlink(DEFAULT_SMGR, userindexRelation) != SM_SUCCESS)
+	if (smgrunlink(DEFAULT_SMGR, userIndexRelation) != SM_SUCCESS)
 		elog(ERROR, "index_destroy: unlink: %m");
 
-	index_close(userindexRelation);
+	/*
+	 * Close rels, but keep locks
+	 */
+	index_close(userIndexRelation);
+	heap_close(userHeapRelation, NoLock);
 
 	RelationForgetRelation(indexId);
 
@@ -1720,6 +1734,30 @@ index_build(Relation heapRelation,
 					 predInfo);
 }
 
+/*
+ * IndexGetRelation: given an index's relation OID, get the OID of the
+ * relation it is an index on.  Uses the system cache.
+ */
+static Oid
+IndexGetRelation(Oid indexId)
+{
+	HeapTuple	tuple;
+	Form_pg_index index;
+
+	tuple = SearchSysCacheTuple(INDEXRELID,
+								ObjectIdGetDatum(indexId),
+								0, 0, 0);
+	if (!HeapTupleIsValid(tuple))
+	{
+		elog(ERROR, "IndexGetRelation: can't find index id %u",
+			 indexId);
+	}
+	index = (Form_pg_index) GETSTRUCT(tuple);
+	Assert(index->indexrelid == indexId);
+
+	return index->indrelid;
+}
+
 /*
  * IndexIsUnique: given an index's relation OID, see if it
  * is unique using the system cache.